Test five things on real work: AI Meeting Notes on a client call, Notion Agent answering questions from your own pages, a Custom Agent on a repeating job, what the AI can and cannot see, and how many credits each test uses. Set a pass mark for each before you start, and upgrade only the people who pass.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowThe trial is where the decision gets made, and it is easy to waste. Teams click around, enjoy a demo of the feature, and upgrade on the strength of the demo. The plan below replaces that with a scored week or two on your own pages, calls and routines, with a limit on what you spend.
What you are actually deciding to pay for
Notion's pricing page lists the Business plan at $20 per member a month when billed yearly (at the time of writing the page also advertises savings of up to 20 per cent for yearly billing, so confirm what monthly billing costs on that page). In the AI column it names Notion Agent, which completes multi-step tasks using context from Notion, connected apps and the web; AI Meeting Notes, which records and summarises meetings without a bot joining the call; and Enterprise Search, marked beta, which searches connected apps such as Slack and GitHub. Custom Agents are described separately: they run on triggers and schedules, and they draw on Notion credits.
The credits matter, because they are where the bill can grow. The Notion pricing and product pages give each Business seat 1,500 Standard credits and 500 Select credits a month. Standard credits cover work where Notion picks the model. Select credits cover work where you pick the model, plus Custom Agents you own. Extra Notion credits are an add-on that workspace owners buy at $10 per 1,000 monthly credits. How those allowances behave is spelled out in the Notion credits explainer, so I only use the essentials here.
Notion's help centre says organisations and teams may be eligible to try the Business plan for 30 days, and that eligibility depends on your workspace. Do not assume you qualify or that the length is the same for you: when you start, find the end date on your own billing screen and work backwards from it.
Set up before day one
- Pick two or three testers. Choose people whose work differs: one who is on client calls, one who writes weekly updates, one who searches for information across projects.
- Write down the current cost of each job. Time spent writing call notes, time spent on the weekly update, time spent hunting for a document. A rough number of minutes is enough, but write it before the trial so you are not guessing afterwards.
- Agree the pass marks. The tests below suggest figures; change them if your business works differently, but do it now.
- Decide what data may be used. Use ordinary internal pages and non-sensitive calls first. If you plan to use client information, read the tutorial on what Notion AI sees and what to restrict before the trial starts.
- Set a credit limit. Workspace owners can set a credit limit in Settings, Notion AI, on the Credits tab; it is off by default. Turn it on so a runaway agent cannot spend freely.
I use a bicycle repair shop with eight staff as the running illustration: one owner, two mechanics who take customer calls, a bookkeeper and four who work the counter. The testers are the owner, one mechanic and the bookkeeper.
The one-paragraph brief for your testers
Testers behave better with a written brief than with a verbal "have a play". Send something like this: "For the next two weeks, use Notion AI only on the five tests in the trial sheet. Log every use in the table with the date, the task and the minutes it took you to fix the result. Do not paste customer names, prices or personal details into any AI request. If you see anything that looks like it came from a page you should not have access to, tell the owner the same day." It takes two minutes to write, and it turns three curious people into three reliable data sources.
Ask each tester for a one-line verdict at the end as well: "Would you be unhappy if this were switched off tomorrow?" That question catches value your scoring misses, and it is the clearest guide to who really needs a seat.
How do you test AI Meeting Notes in a Notion trial?
Notion says AI Meeting Notes needs its Business or Enterprise plan, so the trial is the place to test it on three real calls. You start it on desktop by typing /meet on a page, or from the Meeting notes tab, the compose button or the Upcoming events tile on Home, then choose Start transcribing and Stop at the end. A summary needs at least one minute of audio, and there is a daily usage limit of 10 hours per user.
The desktop app captures your microphone and the system audio, which covers video calls. The browser picks up only your microphone, so it suits in-person meetings. On mobile, the app records through the phone's microphone, which cannot capture other participants on a video call, particularly when headphones are used. Test on the setup you really use for calls, not on a quiet demo.
Run it on three calls of different kinds: a short customer call, a longer supplier call and one with a person who has a strong accent or speaks quickly. Then score each against these questions:
- Did it capture every action item someone said aloud? Count the ones it missed.
- Did it get names, numbers and dates right? Check each against what you remember.
- Did the notes land where the team can find them? Notion saves notes to your default meetings database, or creates a private "My meetings" database if you have none.
- How long did the clean-up take, in minutes, compared with writing notes by hand?
A filled-in scorecard for one illustrative supplier call looks like this:
Call: Supplier call, 22 minutes, desktop app, Zoom
Action items said aloud: 6 Captured: 5 Missed: "order the 26-inch tubes by Friday"
Wrong figures: 1 (quoted 14 units, notes say 40)
Notes saved to: Meetings database (found in 10 seconds)
Clean-up time: 6 minutes Old method: 15 minutes
Verdict: pass on time saved, fail on accuracy until figures are checked by hand
Pass mark: it catches at least 80 per cent of action items on all three calls, with no wrong figure left in the notes after a five-minute check. Consent is part of the test, not an afterthought: Notion's help article suggests a consent message copied into the call chat, played aloud, or simply asked. Try the one you would use with a customer. For a full set of wordings see the AI note-taking consent wording tutorial.
Speaker labelling deserves its own note. Notion says speaker labelling works for English only, though transcription covers sixteen languages. If your calls are in another language, expect transcripts without named speakers.
How do you test Notion Agent in a trial?
Test Notion Agent only on questions whose answers you can already check, because answers you cannot verify prove nothing. Notion Agent completes multi-step tasks using context from Notion, connected apps and the web. So write ten questions whose answers you know and can find in your own workspace: the warranty terms you quote customers, who covers the counter on Fridays, the process for a special order, the price of the standard service.
- Write the ten questions and the correct answers on paper first, with the page each answer lives on.
- Ask Notion Agent each question in a fresh chat, so earlier answers do not colour later ones.
- Mark each answer correct, partly right or wrong. Note whether it cited the right page.
- For every wrong answer, open the source page and decide why. Was the page out of date, contradictory or missing? A wrong answer from a messy page tells you about the page, not the tool.
An illustrative result for the bicycle repair shop: 7 correct, 2 partly right (right policy, wrong year's price) and 1 wrong (it answered from an old duplicate of the returns page). That is a typical pattern, and the fix is housekeeping: archive the old page, retest. The prerequisite work is covered in the Notion knowledge base tutorial.
Pass mark: 8 of 10 correct or partly right, with no answer that would embarrass the business if sent to a customer. Add one trap: ask a question whose answer is not in the workspace. A good result is "I could not find this", not a confident invention.
How do you test a Custom Agent in a trial?
Test one Custom Agent on a weekly repeating job for two weeks, and measure its credit use yourself, because Notion does not publish a credit cost per run. Custom Agents, per Notion's help centre, are background workflows that run on your instructions, on schedules or triggers such as a page added to a database or a Slack message. They need a Business or Enterprise plan, are available to everyone in the workspace by default, and use Notion credits.
Choose a job that repeats weekly, produces text you can check, and does no harm if wrong. A good example is a Monday summary of last week's repair jobs from a database. Avoid agents that send messages to customers during a trial.
- Write the instruction in plain sentences: what to read, what to produce, where to put it, and what to leave out.
- Set a weekly schedule, with a time and time zone.
- Run it manually once to see the output, then let the schedule run for two cycles.
- After each run, note the credits used from the credit dashboard, the accuracy of the content, and the minutes you spent fixing it.
One detail to remember: a Custom Agent you set up uses your Select credits, even when you share it with your team. After those run out it can use extra credits an owner gives you, then workspace Notion credits if an owner allows that. What happens to an agent when credits run out is something to watch for during the trial, so note any run that fails or is skipped; Notion's help article does not say whether it retries. The build is described in the weekly client update tutorial.
Pass mark: two consecutive runs that need under five minutes of correction each, and credit use that the owner could multiply by the number of agents you would really run and still accept.
A sample agent instruction and the output to expect
Vague instructions are the commonest reason a Custom Agent disappoints. Here is the kind of wording to try for the repair shop's Monday summary, followed by an illustrative output and what to fix in it:
Every Monday at 8:00, read the "Repair jobs" database. List jobs completed in the last
7 days, jobs still waiting on parts, and jobs promised for this week. Use three headings
and one line per job: customer first name, bike, status. Do not include prices or phone
numbers. Add the summary to the "Weekly update" page.
Completed (11): Sam, road bike, gear service; Priya, hybrid, brake fit; ...
Waiting on parts (3): Alex, tourer, rear hub; ...
Promised this week (6): Jo, folding bike, Thursday; ...
Note: 2 jobs have no status set.
What to fix: the agent counted 11 completed jobs but one was marked "collected" in a different field, so the true number is 12. The fix is to name that field in the instruction, not to correct the output each week. If a run needs the same correction twice, change the instruction instead.
What can Notion AI see that it should not?
Test what Notion Agent can see by asking it a question whose answer lives only on a page one of your testers cannot open. Access is the test most trials skip. Use, for example, a private owner-only page of staff pay. The right answer is that it cannot find the information. If it reveals it, stop the trial for that area and fix the sharing settings before going further.
- Create a page called "Trial access check" containing a harmless fake fact, such as "The stockroom door code is 0000".
- Restrict the page so only the owner can open it.
- Log in as a tester without access and ask Notion Agent for the door code, in three different wordings.
- Repeat with a connected app, if you use Enterprise Search, using a Slack channel the tester is not in.
Pass mark: no leak, in any of the wordings. This should take ten minutes, and it is the cheapest insurance in the whole trial.
How do you track Notion AI credit use during a trial?
Track credits during the trial with Notion's own credit screens, because credits are the running cost. Members can check their own balances in Settings, Notion AI, under Me. Owners view per-member spending in Settings, Access and billing, Notion credits. Notion Agent also warns when Standard or Select credits are running low. The allowances reset monthly at the start of the billing cycle and unused credits do not roll over, so early use in a billing cycle is not wasted.
Notion's help centre gives general estimates of about 5 credits for a simple task, 10 for a medium one and 25 for a complex one. They are not specific to Custom Agents, so treat them as a rough guide. Keep a log:
| Date | Person | Task | Standard used | Select used | Result |
|---|---|---|---|---|---|
| Day 2 | Owner | Q&A test, 10 questions | about 60 | 0 | 8 of 10 ok |
| Day 4 | Mechanic | Three call summaries | check dashboard | check dashboard | Needs figure check |
| Day 8 | Bookkeeper | Weekly agent, run 1 | 0 | about 25 | Good, 3 minutes fix |
The figures in the first and third rows are illustrative; read yours from the dashboard. What matters is a projection: if the weekly agent costs 25 credits a run, four runs a month is 100 credits, a fifth of the 500 Select credits a seat receives. Five such agents owned by one person would need 500, and that person's Select allowance would be gone. That is the point where extra credits at $10 per 1,000 enter the sums.
If you keep the credit log in a spreadsheet, try The Complete Excel Course with AI Integration on Coding Liquids.
How do you score the trial and decide whether to upgrade?
Score the trial by putting the five results on one page, with each test marked pass, partial or fail against the mark you set in advance, and upgrade only the people who passed.
| Test | Pass mark | Who benefits | Result |
|---|---|---|---|
| 1. Meeting notes | 80% of action items, no wrong figures after check | People on client calls | Pass, with a check step |
| 2. Notion Agent Q&A | 8 of 10 usable, one honest "not found" | Anyone who searches the wiki | Partial, fix duplicates |
| 3. Custom Agent | Two runs, under 5 minutes of fixing | Whoever writes weekly updates | Pass |
| 4. Access | No leak in three wordings | Owner, everyone | Pass |
| 5. Credits | Projected use inside allowances | Owner | Pass for one agent |
The decision rule I would use: upgrade a person if they passed the test that matches their job and test four passed for everyone. Three or more passes for the team means going ahead. Fewer than three means staying where you are, or carrying out the one missing test in a second trial.
Now the sum. If the team is moving from Plus to Business at $20 per member (check Plus on the pricing page; the sum below assumes $10 per member), the extra is $10 per member a month on yearly billing, so $30 a month for the three testers, or $80 for all eight. Compare that with the time saved in the testers' own notes. If the mechanic saves 9 minutes on each of 20 calls a month, that is 3 hours; at an illustrative $25 an hour of staff time, $75 of value against a $10 cost. The same arithmetic with fewer calls and lower hourly cost can easily come out the other way, which is why you measured.
Mistakes that ruin a trial
- Starting on a Friday afternoon. The trial clock runs through the weekend, and your first usable test day is Monday. Start on a Monday or Tuesday.
- Testing the demo, not the job. A tidy practice page proves nothing. Use the real, messy pages.
- Letting every credit-using feature loose at once. You will not know which one used the credits. Stagger the tests.
- No exit plan. Find out before you start what happens to your Custom Agents and AI notes if you decide not to upgrade, and export anything you want to keep.
- Ignoring the end date. Put a reminder two days before the trial ends, with the decision meeting booked, so billing does not decide for you.
If you would like a second pair of eyes on which tests suit your team, that is what my AI implementation consultation covers. For the wider question of whether Notion's AI is worth paying for compared with a general assistant, this tutorial on Notion AI value is the place to continue. Check the live pricing and help pages before you start, because plans and credit rules are Notion's to change.
Notion AI Business trial: what owners ask before upgrading
What should I test during a Notion AI Business trial?
Five things on real work: AI Meeting Notes on a client call, Notion Agent on questions you can check, one Custom Agent on a repeating job, what the AI can and cannot see, and how many credits each test uses. Set a pass mark for each before you start.
How long is the Notion Business trial?
It depends on the offer. Notion's paid-plan trials help page says most trials run for 14 or 30 days, and eligibility depends on your workspace. Look at the end date shown in your own upgrade or billing screen, and put it in the calendar before you start testing.
Do Custom Agents cost extra during the Business plan?
Yes, in general. Notion's pricing page describes Custom Agents as billed through Notion credits at $10 per 1,000 monthly credits, and its help centre says Custom Agents first use the owner's Select credits. Business seats come with allowances, so small tests may cost nothing extra, but check the credit dashboard.
Should every team member join the trial?
No. Test with two or three people who do different work, such as one who runs client calls, one who writes status updates and one who searches the wiki. A small group gives clearer scoring, spends fewer credits and means fewer people disrupted if you decide not to upgrade.
What if the tests are mixed?
Upgrade only the people who passed the tests relevant to their job. Business is billed per member, so a partial rollout is possible. If nobody passes at least three of the five tests, stay on the current plan and revisit in a quarter, when the features and prices will have moved on.
Further reads
- Notion AI Meeting Notes or Fathom for Recording Client Calls? — Compare the meeting-notes result against a free alternative.
- How to Build an AI Operating System for an Ad Agency in Notion — Build a small Notion system for an agency: brand-voice pages, brief templates, agent prompts and approval gates, with a worked brief.
- ClickUp or Notion for a Small Team: Tasks, Docs and AI Cost — A ten-seat cost and workflow test for teams deciding between ClickUp task control and Notion's document-led workspace.
Browse all AI-for-business tutorials
Sources: Notion's pricing page, the Notion help centre articles on AI Meeting Notes, Notion credits, managing AI credits and Custom Agents, and Notion's AI product page, read in November 2026.
